CRM Analytics: Reaching the Heart of the CRM Process

SPEED UP CRM ROI USING BEHAVIORAL METRICS

As reported by Jim Novo, the analytical applications are often one of the last pieces of the CRM puzzle to be implemented. Marketers should be aware there may be additional lag time after installation before these applications are fully operational; machine-learning data mining tools require a "training period". Yet the analytics are key to driving any increase in marketing efficiency and effectiveness! What can the frustrated marketer (and the likely frustrated CEO) do to address this situation?

While you are waiting for the analytics to be selected or installed and trained, use proven and easy to implement proxies for the analytical machines and start to build some experience in marketing directly to customers. When the machines do come online, you will not only be more prepared to better interpret their output, but also have critical behavioral data on your customers to feed into the hungry data hopper.

One of the most powerful and easy to implement of these proxies is tracking customer Recency -- the time elapsed since a customer has engaged in a specific activity with you. The more time elapsing since the customer engaged in an activity, the less likely it becomes the customer will repeat the activity.

There are two reasons Recency is an important "pre-CRM" behavioral metric to work with:

Recency is a powerful behavioral metric all by itself as an indicator of customer loyalty and likelihood to be responsive to promotions.

Recency often becomes a major component of any "discoveries" the machine-based analytical tools find to be important to future customer value and profitability. As long as you have dated customer transaction activity in a database, you can use simple query tools to track customer Recency. By tracking customer Recency, you can get started with high ROI behavior-based customer marketing programs right away, and generate valuable behavioral data to feed the machine-based analytics when they come online.

Recency and Response

Recency has a long history in database marketing, and has proven to be predictive repeatedly across many types of customer behavior. The activity you track for Recency could be purchases, visits, downloads, log-ins -- almost anything that requires an "action" of the part of a customer. Customers engaging in multiple actions could be assigned a Recency metric for each action. For example, the customer could be very Recent on page views but not very Recent on purchases. This would imply the customer is likely to visit again but is becoming less likely to purchase -- just the kind of customer you should make an offer to before they stop coming back to visit.

You will generally see response rates to a promotion asking for a specific action (purchase, visit, click a link) fall as a function of Recency -- the number of weeks or months since the customer last engaged in the activity you are trying to encourage. This relationship is a smooth curve and quite predictable once you establish the "slope" of it for your business. Response rate by Recency might look like this:

• Customer inactive for 1 month, Response rate = 20%
• Customer inactive for 2 months, Response rate = 10%
• Customer inactive for 3 months, Response rate = 4%
• Customer inactive for 4 months, Response rate = 1%

The absolute response rates will be different depending on the business, media used, and offer, but the relative response rates will follow a decelerating curve as shown above, that is, the less Recent the customer, the more dramatic a drop in response rate you will get to your request for an action from the customer.

In terms of using this information for promotions, you will find some point along the curve where you will get "breakeven", meaning the cost of the campaign will equal the profits or benefit generated. For example, let's say you offer a discount, gift, or other incentive in your retention / lapsed customer campaign and need a response rate of at least 4% to pay back the campaign cost. This is your breakeven point.

The implication for this 4% breakeven campaign contained in the Recency information above is this: don't bother to promote to any customer who hasn't engaged in the activity you are trying to encourage for over 3 months, because you're wasting your money. Response will be too low to pay back the cost of the campaign with any customer who has been inactive for over three months.

This Recency effect is very stable over time, allowing you to predict in advance what response to a campaign will be, once you do this "establishing" campaign to see what your response rate is for any particular offer. Recency will predict average response rate for any specific combination of offer and media used. You can save a tremendous amount of money by forecasting your response by using Recency, and not promoting to customers unlikely to be profitable.

Actionable Item: Set up and execute an establishing Recency test.

Classify customers in 30-day Recency segments by the last date of the activity you want to profile for Recency. If you want to profile purchases, customers could be segmented by date of last purchase:

In the past 30 days

31 - 60 days ago

61 - 90 days ago

91 - 120 days ago

121 - 150 days ago

151 - 180 days ago

181+ days ago

Take a 10% random sample of customers from each segment (every 10th person in the segment), and send all of them a promotion with the same offer, say 20% off any purchase in the next 30 days. Look at the response rate by these 30-day segments. You will find response falls off significantly as you look at Recency segments further back in time. If you repeat the test using the same offer to a different sample of each 30-day segment, the response rate by segment will be very close to the response rate by segment in the first test. This kind of stability allows accurate predictions of marketing ROI before promotions are even sent out to customers.

Recency and Offers

The response rate in any one of the 30-day segments above will be influenced by the value of your offer, and both response rate and the cost of the offer have significant impact on the profitability of your campaign to any segment.

As offer value increases, so does response rate, and so do costs. Ideally, you want to find the ideal mix of response rate and offer value creating the highest profitability for each segment you promote to.

You can use Recency to "ladder" the promotional discount, gift, or incentive value offered in a promotion, boosting overall response while cutting expenses by minimizing discount or other incentive costs.

Let's use purchases as an example, and say you usually e-mail all your customers a 10% discount when you do a promotion. If you were using a Recency ladder approach for this purchase incentive, you might apply your discount strategy this way:

• Customer inactive for 1 month, Response rate = 20%, discount = 5%
• Customer inactive for 2 months, Response rate = 10%, discount = 10%
• Customer inactive for 3 months, Response rate = 4%, discount = 15%
• Customer inactive for = 4 months, Response rate = 1%, discount = 20%

Using this approach, you are allocating the most "bang for the buck" discount-wise where you need it most -- the least Recent, lowest response customers, and pulling back on some discounting where you don't need it as much -- the most Recent, highest response customers.

Since your most Recent customers are most likely to respond, you can back off on their discount and you reduce the cost of giving discounts to customers who "may have bought anyway without a discount". You then reallocate this discount money to where it is needed most -- boosting the response rates of those less likely to respond -- less Recent customers.

Your response rates will vary depending on the offer, media used, and your business. You have to test these ladders with different combinations of offer and media to find the optimum profitability for each Recency segment. The interesting and quite useful benefit of this approach is the "automatic" overall customer retention effect discount ladders have.

Using a ladder of this type means your promotional discount budget is automatically working harder and harder to keep a customer active with you as they drift further and further away from you. The less Recent a customer is, the less likely they are to buy or visit again, and by using a discount ladder you are counteracting the customer LifeCycle (the tendency of customers to leave you over time) with stronger discounts as the defecting customer behavior plays out.

If a most Recent customer does not respond to the 5% offer, as they get less Recent, they automatically get offers rising in value, and at some point, many will take advantage of an offer. The customers who run through this system without taking any offers were likely lost to you as a customer already, and not worth the extra expense to try and keep promoting to them.

Actionable Item: Set up and execute a discount ladder test.

Pick any one of the segments from your Recency test above and now test discount level for the segment. Let's say you used a 20% discount in the first test. Pick a segment (say 91 - 120 days), and create a 20% random sample of the segment (every 5th customer) divided into 4 equal test groups. Send each test group a different discount -- say 5%, 10%, 15% and 20%. Look at your response rates and calculate the profitability for the 91 - 120 day segment at each discount level.

You will find your result looks similar to the following table:

Discount Test : 91 - 120 Day Recency Segment

Customer Sample 1000 1000 1000 1000

Discount Offer 5% 10% 15% 20%

Response Rate 2% 4% 6% 8%

Responders 20 40 60 80

Average Price $80 $80 $80 $80

Totals Sales $1,600 $3,200 $4,800 $6,400

Gross Margin 30% 30% 30% 30%

Gross Profit $480 $960 $1,440 $1,920

Discount Cost $80 $320 $720 $1,280

Net Cost Before

Media Expense $400 $640 $720 $640

As you can see, the most profitable offer to the 91 - 120 day Recency segment is 15% off. If you offer 20%, you get a higher response rate but lower profits; any offer under 15% significantly diminishes response rate. Repeat this test for each Recency segment, and you will find the most profitable discount rising as the customer becomes less Recent, creating your discount "ladder".

When you implement your promotions based on this ladder, as customers become less Recent and therefore less likely to respond to a promotion, they will be automatically offered a higher discount -- one that maximizes profit for each Recency segment the customer passes through.

Conclusion

The data points you gather from determining how Recency affects response and offer are extremely valuable to the data mining effort because they are based on actual customer behavior rather than "softer" attributes like demographics. The basic parameters of Recency, response, and offer represent actual bottom-line financial impact to the enterprise. When you include variables with a known financial impact in your data mining process, the predictions and correlations output by the miners can actually be used to increase the profitability of the business.

This approach to creating a customer retention program is clean, simple, and easy to implement. And if you don't have any formal customer retention program in place, much better than what you're using now!

Jim Novo is an interactive retailing expert with over 15 years of experience generating high ROI customer marketing programs for Home Shopping Network, CBS/SportsLine, MBNA, and many others. Jim's book, Drilling Down: Turning Customer Data into Profits with a Spreadsheet, describes simple methods for using customer data to create high ROI customer marketing programs. More "How To" customer marketing articles can be found on his Website: www.jimnovo.com.